M.S. Applied Data Science - Capstone Chronicles 2025

5

will produce a business-ready output with top strategic targets for cross-promotion efforts and an accompanying map for visualizing graph edges, communities, and additional node details. Literature Review Understanding the intricate web of relationships between brands is a central goal of modern marketing strategy. Historically, this “market structure” was mapped using survey data and consumer panels. However, the availability of large-scale transactional data and advancements in machine learning have provided novel, data-driven approaches to identify latent market structures. While previous research focused on the use of unsupervised machine learning to identify market segments using various clustering algorithms, contemporary approaches increasingly attempt to capture the directional and networked nature of relationships between products and brands. A Latent-Class Probabilistic Model for Market Structure Analysis Statistical approaches, such as latent-class analysis, have been employed to effectively identify subgroups in competitive market structures (Grover & Srinivasan, 1987). In the study, Grover and Srinivasan proposed a latent-class probabilistic model using consumer brand switching data to infer competitive market structures and identify customer segments simultaneously. Their model accounts for consumer heterogeneity by postulating that market segments correspond to latent consumer classes, where competitive relationships were revealed by brand choice probabilities in each class. The data utilized in this study aggregates cross shopping spend data to a physical retail location; we aim to build upon this research by inferring competitive and complementary market structures revealed by brand choices at the transaction level.

213